CODEWITHSUNDEEP
raggregatelinear-regression
hannsta
hannsta

Reputation: 109

Aggregate linear regression

Sorry I am quite new to R, but I have a dataframe with gamelogs for multiple players. I am trying to get the slope coefficient for each player's points over all of their games. I have seen that aggregate can use operators like sum and average, and getting coefficients off of a linear regression is pretty simple as well . How do I combine these?

a <- c("player1","player1","player1","player2","player2","player2")
b <- c(1,2,3,4,5,6)
c <- c(15,12,13,4,15,9)
gamelogs <- data.frame(name=a, game=b, pts=c)

I want this to become:

   name    pts slope
player1       -.4286
player2       .08242

Upvotes: 3

Views: 2925

Answers (3)

thelatemail
thelatemail

Reputation: 93813

You can also do some magic with the base lm to do it all at once:

coef(lm(game ~ pts*name - pts, data=gamelogs))[3:4]
coef(lm(game ~ pts:name + name, data=gamelogs))[3:4]
#pts:nameplayer1 pts:nameplayer2 
#    -0.42857143      0.08241758

As a data.frame:

data.frame(slope=coef(lm(game ~ pts*name - pts, data=gamelogs))[3:4])
#                      slope
#pts:nameplayer1 -0.42857143
#pts:nameplayer2  0.08241758

See here for some further explanation of the modelling in the lm call:

https://stat.ethz.ch/R-manual/R-devel/library/stats/html/formula.html
http://faculty.chicagobooth.edu/richard.hahn/teaching/FormulaNotation.pdf#2

In this case pts*name expands to pts + name + pts:name which when removing - pts means it is equivalent to pts:name + name

Upvotes: 6

Rich Scriven
Rich Scriven

Reputation: 99331

You could do

s <- split(gamelogs, gamelogs$name)

vapply(s, function(x) lm(game ~ pts, x)[[1]][2], 1)
#     player1     player2 
# -0.42857143  0.08241758

or

do.call(rbind, lapply(s, function(x) coef(lm(game ~ pts, x))[2]))
#                 pts
# player1 -0.42857143
# player2  0.08241758

Or if you want to use dplyr, you can do

library(dplyr)

models <- group_by(gamelogs, name) %>% 
    do(mod = lm(game ~ pts, data = .))

cbind(
    name = models$name, 
    do(models, data.frame(slope = coef(.$mod)[2]))
)
#      name       slope
# 1 player1 -0.42857143
# 2 player2  0.08241758

Upvotes: 5

Rorschach
Rorschach

Reputation: 32426

library nlme has a function for this as well, lmList

library(nlme)
coef(lmList(game ~ pts | name, gamelogs))
#        (Intercept)         pts
# player1    7.714286 -0.42857143
# player2    4.230769  0.08241758

Upvotes: 4

Related Questions